Enable stand-alone tiling, but it gives a warning and converting to simd.

Author: jsjodin

flang/lib/Lower/OpenMP/OpenMP.cpp

@@ -2169,6 +2169,40 @@ genLoopOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
   return loopOp;
 }
 
+static mlir::omp::LoopOp
+genTiledLoopOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
+          semantics::SemanticsContext &semaCtx, lower::pft::Evaluation &eval,
+          mlir::Location loc, const ConstructQueue &queue,
+          ConstructQueue::const_iterator item) {
+  mlir::omp::LoopOperands loopClauseOps;
+  llvm::SmallVector<const semantics::Symbol *> loopReductionSyms;
+  genLoopClauses(converter, semaCtx, item->clauses, loc, loopClauseOps,
+                 loopReductionSyms);
+
+  DataSharingProcessor dsp(converter, semaCtx, item->clauses, eval,
+                           /*shouldCollectPreDeterminedSymbols=*/true,
+                           /*useDelayedPrivatization=*/true, symTable);
+  dsp.processStep1(&loopClauseOps);
+
+  mlir::omp::LoopNestOperands loopNestClauseOps;
+  llvm::SmallVector<const semantics::Symbol *> iv;
+  genLoopNestClauses(converter, semaCtx, eval, item->clauses, loc,
+                     loopNestClauseOps, iv);
+
+  EntryBlockArgs loopArgs;
+  loopArgs.priv.syms = dsp.getDelayedPrivSymbols();
+  loopArgs.priv.vars = loopClauseOps.privateVars;
+  loopArgs.reduction.syms = loopReductionSyms;
+  loopArgs.reduction.vars = loopClauseOps.reductionVars;
+
+  auto loopOp =
+      genWrapperOp<mlir::omp::LoopOp>(converter, loc, loopClauseOps, loopArgs);
+  genLoopNestOp(converter, symTable, semaCtx, eval, loc, queue, item,
+                loopNestClauseOps, iv, {{loopOp, loopArgs}},
+                llvm::omp::Directive::OMPD_loop, dsp);
+  return loopOp;
+}
+
 static mlir::omp::MaskedOp
 genMaskedOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
             lower::StatementContext &stmtCtx,
@@ -3648,6 +3682,9 @@ static void genOMPDispatch(lower::AbstractConverter &converter,
                        item);
     break;
   case llvm::omp::Directive::OMPD_tile:
+    newOp =
+      genTiledLoopOp(converter, symTable, semaCtx, eval, loc, queue, item);
+    break;
   case llvm::omp::Directive::OMPD_unroll: {
     unsigned version = semaCtx.langOptions().OpenMPVersion;
     TODO(loc, "Unhandled loop directive (" +

flang/test/Lower/OpenMP/wsloop-collapse.f90

@@ -57,7 +57,7 @@ program wsloop_collapse
 !CHECK:           %[[VAL_31:.*]] = fir.load %[[VAL_11]]#0 : !fir.ref<i32>
 !CHECK:           %[[VAL_32:.*]] = arith.constant 1 : i32
 !CHECK:           omp.wsloop private(@{{.*}} %{{.*}}#0 -> %[[VAL_4:.*]], @{{.*}} %{{.*}}#0 -> %[[VAL_2:.*]], @{{.*}} %{{.*}}#0 -> %[[VAL_0:.*]] : !fir.ref<i32>, !fir.ref<i32>, !fir.ref<i32>) {
-!CHECK-NEXT:        omp.loop_nest (%[[VAL_33:.*]], %[[VAL_34:.*]], %[[VAL_35:.*]]) : i32 = (%[[VAL_24]], %[[VAL_27]], %[[VAL_30]]) to (%[[VAL_25]], %[[VAL_28]], %[[VAL_31]]) inclusive step (%[[VAL_26]], %[[VAL_29]], %[[VAL_32]]) {
+!CHECK-NEXT:        omp.loop_nest (%[[VAL_33:.*]], %[[VAL_34:.*]], %[[VAL_35:.*]]) : i32 = (%[[VAL_24]], %[[VAL_27]], %[[VAL_30]]) to (%[[VAL_25]], %[[VAL_28]], %[[VAL_31]]) inclusive step (%[[VAL_26]], %[[VAL_29]], %[[VAL_32]]) collapse(3) {
   !$omp do collapse(3)
   do i = 1, a
      do j= 1, b